Consistency diffusion language models: Up to 14x faster, no quality loss

by zagwdt | View on Hacker News